JP2819722B2 - Tone generator - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Industrial Field of the Invention The present invention relates to a tone generator having a sound source circuit such as a so-called FM sound source.

(B) Conventional technology Conventionally, FM sound sources have been put to practical use. The FM sound source synthesizes a plurality of (4 or 6) operators that can independently set parameters such as a basic waveform (sine wave, triangular wave, etc.), frequency (pitch), output level, envelope, etc. based on a specific algorithm ( (Modulation and addition) to generate one musical sound. This sound source can generate a variety of beautiful musical sounds with a simple configuration.

(C) Problems to be Solved by the Invention However, it has been considered that it is difficult for this FM sound source to generate a timbre as intended by the author. This is mainly
Since musical tones are generated using many parameters and FM modulation, it is difficult to imagine how musical tones change with changes in operators and algorithms.

An object of the present invention is to provide a tone generator capable of easily generating a tone according to an image by using a neural network that infers and outputs a waveform determination parameter based on a plurality of image parameters representing characteristics of a tone. With the goal.

(D) Means for Solving the Problems The present invention provides a tone generator circuit that generates an operator based on a plurality of waveform determination parameters and synthesizes the generated plurality of operators with a predetermined algorithm to generate one musical tone. Means for inputting a plurality of image parameters representing characteristics of timbres; a neural network for inferring and outputting the waveform determining parameters and the algorithm based on the input plurality of image parameters; Means for setting the waveform determination parameters and the algorithm inferred and output by the network in the tone generator circuit.

(E) Effects of the Invention The musical sound generating device of the present invention generates musical sounds by synthesizing a plurality of operators generated based on a plurality of waveform determination parameters. The neural network infers and outputs the waveform determination parameters and the synthesis algorithm for generating the musical sound. The neural network is trained in advance so that the parameters of the musical tone imaged by a general user can be output. As input data for inference, image parameters expressing characteristics of musical sounds are used. The image parameters include, for example, the hardness / softness of the sound, the loudness / soberness of the sound, the beauty of the sound /
Parameters such as dirtyness and sound thickness / thinness may be used. Thus, by inputting an image, a musical tone suitable for the image can be generated.

(F) Embodiment FIG. 1 is a block diagram of a control unit of an electronic musical instrument according to an embodiment of the present invention. The electronic musical instrument is controlled by the CPU 10 and generates and generates musical tones based on the player's operation. The CPU 10 is connected to each circuit via a bus 11. The circuit section includes ROM 12, RAM 13, neural network (NN) 14, keyboard 15, operation panel 16, FM sound source circuit 1
There are seven. A sound system 18 for amplifying the generated musical sound and outputting the amplified musical sound from a speaker or the like is connected to the FM sound source circuit 17.

The ROM 12 stores programs, synapse weights for presets, and the like.The RAM 13 is provided with registers for storing various data generated during the performance. The weight is stored.

Further, the neural network 14 has a function of determining a waveform determination parameter or the like in the FM sound source based on the input image parameters. The neural network 14 is a hierarchical neural network as shown in a schematic diagram in FIG. 2, and image parameters are inputted to the input layer from the tone color designation dials 16a to 16d, and the output layer is inferred based on these parameters. Outputs waveform determination parameters as shown in FIG. This neural network may have any hardware configuration as long as the illustrated hierarchical inference is possible. The keyboard 15 is for performance and has a range of about 5 octaves. The operation panel 16 is provided with a learning mode / normal mode selection switch, a synapse weight selection switch, and the like, in addition to the tone color image designation dials 16a to 16d and the like. Learning mode is
In this mode, a tone color learning is performed by inputting an image of a waveform determination parameter or the like set at that time using a tone color image designation dial. The normal mode is a normal performance mode. The synapse weight selection switch is a switch for selecting whether to use the synapse weight stored in the ROM 12 in advance or to use the synapse weight learned in the learning mode and stored in the RAM 13.

The FM tone generator circuit 17 determines the contents generated by each of the operators 4 or 6 by setting a plurality of parameters, and synthesizes these operators by a designated algorithm (= combination of operators, modulation). This is a circuit that generates musical tones. By appropriately adjusting each parameter and algorithm, a complex tone change and higher harmonics can be obtained. Each parameter or algorithm is set by the CPU 10 before the performance, and generates a musical tone of the pitch based on a key-on signal or key code sent from the CPU 10 during the performance.

FIG. 2 shows a schematic diagram of the tone image designation dials 16a to 16d and a schematic configuration diagram of the neural network 14. The tone image designation dials 16a to 16d are used to set the degree of the tone image for the four types of tone images, 16a is the hardness of the sound (hard / soft), 16b is the thickness of the sound (thick / thin), 16c is a dial to specify the degree of sound beauty (beautiful / dirty), and 16d is a dial to specify the degree of flashiness (flashy / quiet). The values designated by these tone image designation dials 16a to 16d are input to the neural network 14 as image parameters. These four image parameters are input to the input layer of the neural network 14. Each neuron in the input layer → association layer is synapse-coupled with a predetermined weight, and each neuron in the association layer → output layer is also synapse-coupled with a predetermined weight. The output of each neuron in the output layer corresponds to the waveform determination parameters and algorithm of each operator. For simplicity of explanation, only ON / OFF, frequency ratio (pitch with respect to frequency of generated musical tone), and envelope plate are shown as waveform determination parameters for each operator. Is determined by the following parameters. This output is
To the FM sound source circuit 17, and the FM sound source circuit 17 generates a musical tone based on this parameter.

This neural network 14 has been learned in advance by a plurality of teacher data, and based on this learning, whatever parameters are input, it is possible to determine appropriate output parameters corresponding to the input parameters. The teacher data uses statistical data of an unspecified number of authors, and outputs musical tones according to the image specified by the tone image specification dial 16 and waveform determination parameters generated by the FM tone generator circuit 17. Have been learned to. This greatly reduces the burden on sound creation.

In the above embodiment, the learned synapse weights stored in the RAM 13 may be a plurality of sets.
The musical tone data to be learned may be data input from another device instead of the waveform determining parameters set in the electronic musical instrument.

(G) Effect of the Invention As described above, the musical sound generating apparatus of the present invention can generate musical sounds using a neural network, and can easily generate musical sounds that match an image without operating complicated parameters. can do. As a result, for example, tone editing of the FM tone generator circuit can be facilitated.

[Brief description of the drawings]

FIG. 1 is a block diagram of a control unit of an electronic musical instrument according to an embodiment of the present invention, and FIG. 2 is a diagram showing a configuration of a neural network used in the control unit. 14 ... Neural network, 16a-16d ... Tone image setting dial.

Claims (1)

(57) [Claims] 1. A tone generator comprising a tone generator circuit for generating an operator based on a plurality of waveform determination parameters and synthesizing the generated plurality of operators by a predetermined algorithm to generate one tone. Means for inputting a plurality of image parameters representing features; a neural network for inferring and outputting the waveform determining parameters and the algorithm based on the input plurality of image parameters; a waveform determining parameter inferred and output by the neural network; Means for setting an algorithm in the tone generator circuit.